[llvm] [InstCombine] Remove scalable vector extracts to and from the same type (PR #69702)

Kerry McLaughlin via llvm-commits llvm-commits at lists.llvm.org
Fri Oct 20 03:01:09 PDT 2023


https://github.com/kmclaughlin-arm created https://github.com/llvm/llvm-project/pull/69702

visitCallInst already looks for fixed width vector extracts where number of elements in
the source and destination types are equal. This patch modifies the function to also
identify scalable extracts which can be removed.

>From 8ee1ab40ffb696178777cbb30d66be8aa0058025 Mon Sep 17 00:00:00 2001
From: Kerry McLaughlin <kerry.mclaughlin at arm.com>
Date: Thu, 19 Oct 2023 16:12:10 +0000
Subject: [PATCH] [InstCombine] Remove scalable vector extracts to and from the
 same type

visitCallInst already looks for fixed width vector extracts where the
source and destination types are equal. This patch modifies the function
to also identify scalable extracts which can be removed.
---
 .../InstCombine/InstCombineCalls.cpp           | 18 ++++++++++--------
 .../InstCombine/canonicalize-vector-extract.ll | 10 ++++++++++
 2 files changed, 20 insertions(+), 8 deletions(-)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index e29fb869686ca0b..00505b5edb58dca 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -2997,24 +2997,26 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
         return replaceOperand(CI, 0, InsertTuple);
     }
 
-    auto *DstTy = dyn_cast<FixedVectorType>(ReturnType);
-    auto *VecTy = dyn_cast<FixedVectorType>(Vec->getType());
+    auto *DstTy = dyn_cast<VectorType>(ReturnType);
+    auto *VecTy = dyn_cast<VectorType>(Vec->getType());
 
-    // Only canonicalize if the destination vector and Vec are fixed
-    // vectors.
     if (DstTy && VecTy) {
-      unsigned DstNumElts = DstTy->getNumElements();
-      unsigned VecNumElts = VecTy->getNumElements();
+      auto DstEltCnt = DstTy->getElementCount();
       unsigned IdxN = cast<ConstantInt>(Idx)->getZExtValue();
 
       // Extracting the entirety of Vec is a nop.
-      if (VecNumElts == DstNumElts) {
+      if (DstEltCnt == VecTy->getElementCount()) {
         replaceInstUsesWith(CI, Vec);
         return eraseInstFromFunction(CI);
       }
 
+      // Only canonicalize to shufflevector if the destination vector and
+      // Vec are fixed vectors.
+      if (dyn_cast<ScalableVectorType>(VecTy) || DstEltCnt.isScalable())
+        break;
+
       SmallVector<int, 8> Mask;
-      for (unsigned i = 0; i != DstNumElts; ++i)
+      for (unsigned i = 0; i != DstEltCnt.getKnownMinValue(); ++i)
         Mask.push_back(IdxN + i);
 
       Value *Shuffle = Builder.CreateShuffleVector(Vec, Mask);
diff --git a/llvm/test/Transforms/InstCombine/canonicalize-vector-extract.ll b/llvm/test/Transforms/InstCombine/canonicalize-vector-extract.ll
index 26af633698f3552..0b7f1ebc040c2a1 100644
--- a/llvm/test/Transforms/InstCombine/canonicalize-vector-extract.ll
+++ b/llvm/test/Transforms/InstCombine/canonicalize-vector-extract.ll
@@ -10,6 +10,7 @@ declare <3 x i32> @llvm.vector.extract.v3i32.v8i32(<8 x i32> %vec, i64 %idx)
 declare <4 x i32> @llvm.vector.extract.v4i32.nxv4i32(<vscale x 4 x i32> %vec, i64 %idx)
 declare <4 x i32> @llvm.vector.extract.v4i32.v8i32(<8 x i32> %vec, i64 %idx)
 declare <8 x i32> @llvm.vector.extract.v8i32.v8i32(<8 x i32> %vec, i64 %idx)
+declare <vscale x 8 x i32> @llvm.vector.extract.nxv8i32.nxv8i32(<vscale x 8 x i32> %vec, i64 %idx)
 
 ; ============================================================================ ;
 ; Trivial cases
@@ -24,6 +25,15 @@ define <8 x i32> @trivial_nop(<8 x i32> %vec) {
   ret <8 x i32> %1
 }
 
+define <vscale x 8 x i32> @trivial_nop_scalable(<vscale x 8 x i32> %vec) {
+; CHECK-LABEL: define <vscale x 8 x i32> @trivial_nop_scalable(
+; CHECK-SAME: <vscale x 8 x i32> [[VEC:%.*]]) {
+; CHECK-NEXT:    ret <vscale x 8 x i32> [[VEC]]
+;
+  %ext = call <vscale x 8 x i32> @llvm.vector.extract.nxv8i32.nxv8i32(<vscale x 8 x i32> %vec, i64 0)
+  ret <vscale x 8 x i32> %ext
+}
+
 ; ============================================================================ ;
 ; Valid canonicalizations
 ; ============================================================================ ;



More information about the llvm-commits mailing list